Remaining Useful Life Prediction of Mechanical Components based on CNN-Transformer Parallel Fusion Model

Aiming at the problem that the remaining useful life prediction method of mechanical components is not fully utilized in local degradation features and long-distance time-series dependence, a remaining useful life prediction method based on a parallel fusion model of CNN-Transformer is proposed. Firstly, the time-domain, frequency-domain and time-domain degradation sensitive features are extracted from the vibration signal, and the sliding time window is used to construct multidimensional time series samples; Secondly, the convolutional neural network （CNN） module is used to extract local degradation features, and the Transformer encoder module is used to model the long-distance time-series dependence; Finally, the two kinds of features are fused in parallel, and the remaining useful life is predicted through the output layer. The experimental results of FEMTO-ST bearing data sets show that the proposed model achieves high prediction accuracy under different working conditions, and the R² is more than 0.99, and the RMSE, MAE and R² are better than the single CNN and Transformer model.